[jboss-user] [JBossCache] - Re: Problems with NodeLocking algorithm

[manik.surtani@jboss.com]

First off, thanks for all the effort, Jacek.  

The race condition between the locking and checking of node existence is a nasty bug.  Adding a timer within the loop to break after a timeout is valuable too.

All this really stems from the fact that a Node owns the lock.  This design flaw is being worked on (new MVCC locking design deals with a striped lock on the Fqn) so even if new nodes are created or destroyed, as long as they belong to the same Fqn the lock will apply.

I'm looking into the patches you sent me - I will definitely be using some of them in 1.4.1.SP5.

Regarding your scenario 3 though, I would try RepeatableRead and see if it really does impact your performance all that much.



View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4084417#4084417

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4084417